# Are transposed convolutions computed using the Fast Fourier Transform?

In convolutional neural networks, transposed convolutions are represented by the transposed matrix which represents the convolution. But convolutions are not actually computed by creating this matrix, they are rather computed efficiently by applying the Fast Fourier Transform and multiplying in the Fourier domain. Are transposed convolutions also computed in the Fourier domain? If so, what is the exact operation that is used in the Fourier domain?

• Do you have elements of answer now? I'm also interested! Apr 16 '21 at 7:38